AI/rocm-systems
2
0
Forka 0
Codice Problemi Pull Request Actions Pacchetti Progetti Rilasci Wiki Attività

Add numPipes built-in-var for updated metrics

Sfoglia il codice sorgente 
Co-authored-by: Nick Curtis <nicholas.curtis@amd.com>
Signed-off-by: colramos-amd <colramos@amd.com>


[ROCm/rocprofiler-compute commit: d16f58b797]
This commit is contained in:
colramos-amd
2024-01-31 18:25:37 -06:00
committato da Cole Ramos
parent ef60d67744
commit bd22355b96
7 ha cambiato i file con 44 aggiunte e 29 eliminazioni
Scarica il file Patch Scarica il file Diff Expand all files Collapse all files
projects/rocprofiler-compute/src/omniperf_soc/soc_gfx906.py
+1
Vedi File
@@ -31,6 +31,7 @@ SOC_PARAM = {
"numSE": 4,
"numCU": 60,
"numSIMD": 240,
"numPipes": 4,
"numWavesPerCU": 40,
"numSQC": 15,
"L2Banks": 16,
projects/rocprofiler-compute/src/omniperf_soc/soc_gfx908.py
+1
Vedi File
@@ -31,6 +31,7 @@ SOC_PARAM = {
"numSE": 8,
"numCU": 120,
"numSIMD": 480,
"numPipes": 4,
"numWavesPerCU": 40,
"numSQC": 30,
"L2Banks": 32,
projects/rocprofiler-compute/src/omniperf_soc/soc_gfx90a.py
+1
Vedi File
@@ -33,6 +33,7 @@ SOC_PARAM = {
"numSE": 8,
"numCU": 110,
"numSIMD": 440,
"numPipes": 4,
"numWavesPerCU": 32,
"numSQC": 56,
"L2Banks": 32,
projects/rocprofiler-compute/src/utils/parser.py
+17 -11
Vedi File
@@ -116,6 +116,11 @@ def to_min(*args):
def to_max(*args):
if len(args) == 1 and isinstance(args[0], pd.core.series.Series):
return args[0].max()
elif len(args) == 2 and (
isinstance(args[0], pd.core.series.Series)
or isinstance(args[1], pd.core.series.Series)
):
return np.maximum(args[0], args[1])
elif max(args) == None:
return np.nan
else:
@@ -303,7 +308,7 @@ def build_eval_string(equation, coll_level):
def update_denom_string(equation, unit):
"""
Update $denom in equation with runtime nomorlization unit.
Update $denom in equation with runtime normalization unit.
"""
if not equation:
return ""
@@ -318,7 +323,7 @@ def update_denom_string(equation, unit):
def update_normUnit_string(equation, unit):
"""
Update $normUnit in equation with runtime nomorlization unit.
Update $normUnit in equation with runtime normalization unit.
It is string replacement for display only.
"""
@@ -327,8 +332,8 @@ def update_normUnit_string(equation, unit):
return ""
return re.sub(
"\((?P<PREFIX>\w*)\s+\+\s+(\$normUnit\))",
"\g<PREFIX> " + re.sub("_", " ", unit),
r"\((?P<PREFIX>\w*)\s+\+\s+(\$normUnit\))",
r"\g<PREFIX> " + re.sub("_", " ", unit),
str(equation),
).capitalize()
@@ -690,9 +695,10 @@ def eval_metric(dfs, dfs_type, sys_info, soc_spec, raw_pmc_df, debug):
# NB:
# Following with Omniperf 0.2.0, we are using HW spec from sys_info instead.
# The soc_spec is not in using right now, but can be used to do verification
# aganist sys_info, forced theoretical evaluation, or supporting tool-chains
# against sys_info, forced theoretical evaluation, or supporting tool-chains
# broken.
ammolite__numSE = sys_info.numSE
ammolite__numPipes = sys_info.numPipes
ammolite__numCU = sys_info.numCU
ammolite__numSIMD = sys_info.numSIMD
ammolite__numWavesPerCU = sys_info.maxWavesPerCU # todo: check do we still need it
@@ -739,7 +745,7 @@ def eval_metric(dfs, dfs_type, sys_info, soc_spec, raw_pmc_df, debug):
print("~" * 40 + "\nExpression:")
print(expr, "=", row[expr])
print("Inputs:")
matched_vars = re.findall("ammolite__\w+", row[expr])
matched_vars = re.findall(r"ammolite__\w+", row[expr])
if matched_vars:
for v in matched_vars:
print(
@@ -749,12 +755,12 @@ def eval_metric(dfs, dfs_type, sys_info, soc_spec, raw_pmc_df, debug):
eval(compile(v, "<string>", "eval")),
)
matched_cols = re.findall(
"raw_pmc_df\['\w+'\]\['\w+'\]", row[expr]
r"raw_pmc_df\['\w+'\]\['\w+'\]", row[expr]
)
if matched_cols:
for c in matched_cols:
m = re.match(
"raw_pmc_df\['(\w+)'\]\['(\w+)'\]", c
r"raw_pmc_df\['(\w+)'\]\['(\w+)'\]", c
)
t = raw_pmc_df[m.group(1)][
m.group(2)
@@ -778,7 +784,7 @@ def eval_metric(dfs, dfs_type, sys_info, soc_spec, raw_pmc_df, debug):
print("~" * 40)
except TypeError:
print(
"skiping entry. Encounterd a missing counter"
"skipping entry. Encounterd a missing counter"
)
print(expr, " has been assigned to None")
print(np.nan)
@@ -788,7 +794,7 @@ def eval_metric(dfs, dfs_type, sys_info, soc_spec, raw_pmc_df, debug):
== "'NoneType' object has no attribute 'get'"
):
print(
"skiping entry. Encounterd a missing csv"
"skipping entry. Encounterd a missing csv"
)
print(np.nan)
else:
@@ -894,7 +900,7 @@ def apply_filters(workload, dir, is_gui, debug):
print("{} is an invalid dispatch id.".format(d))
sys.exit(1)
if ">" in workload.filter_dispatch_ids[0]:
m = re.match("\> (\d+)", workload.filter_dispatch_ids[0])
m = re.match(r"\> (\d+)", workload.filter_dispatch_ids[0])
ret_df = ret_df[
ret_df[schema.pmc_perf_file_prefix]["Dispatch_ID"] > int(m.group(1))
]
projects/rocprofiler-compute/src/utils/schema.py
+17 -17
Vedi File
@@ -79,7 +79,7 @@ supported_field = [
"Min",
"Max",
"Avg",
"PoP",
"Pct of Peak",
"Peak",
"Count",
"Mean",
@@ -92,22 +92,22 @@ supported_field = [
"Channel",
"L2 Cache Hit Rate",
"Requests",
"L1-L2 Read",
"L1-L2 Write",
"L1-L2 Atomic",
"L2-EA Read",
"L2-EA Write",
"L2-EA Atomic",
"L2-EA Read Latency",
"L2-EA Write Latency",
"L2-EA Atomic Latency",
"L2-EA Read Stall - IO",
"L2-EA Read Stall - GMI",
"L2-EA Read Stall - DRAM",
"L2-EA Write Stall - IO",
"L2-EA Write Stall - GMI",
"L2-EA Write Stall - DRAM",
"L2-EA Write Stall - Starve",
"L2 Read Req",
"L2 Write Req",
"L2 Atomic Req",
"L2 - Fabric Read Req",
"L2 - Fabric Write and Atomic Req",
"L2 - Fabric Atomic Req",
"L2 - Fabric Read Latency",
"L2 - Fabric Write Latency",
"L2 - Fabric Atomic Latency",
"L2 - Fabric Read Stall (PCIe)",
"L2 - Fabric Read Stall (Infinity Fabric™)",
"L2 - Fabric Read Stall (HBM)",
"L2 - Fabric Write Stall (PCIe)",
"L2 - Fabric Write Stall (Infinity Fabric™)",
"L2 - Fabric Write Stall (HBM)",
"L2 - Fabric Write Starve",
]
# The prefix of raw pmc_perf.csv
projects/rocprofiler-compute/src/utils/specs.py
+5
Vedi File
@@ -63,6 +63,7 @@ class MachineSpecs:
L2Banks: str
LDSBanks: str
numSQC: str
numPipes: str
hbmBW: str
compute_partition: str
memory_partition: str
@@ -96,6 +97,7 @@ class MachineSpecs:
L2Banks: {self.L2Banks}
LDSBanks: {self.LDSBanks}
numSQC: {self.numSQC}
numPipes: {self.numPipes}
hbmBW: {self.hbmBW} MB/s
compute_partition: {self.compute_partition}
memory_partition: {self.memory_partition}
@@ -114,6 +116,7 @@ def gpuinfo():
"max_sclk": None,
"num_CU": None,
"num_SIMD": None,
"numPipes": None,
"num_SE": None,
"wave_size": None,
"grp_size": None,
@@ -197,6 +200,7 @@ def gpuinfo():
gpu_info['L2Banks'] = str(soc_module.SOC_PARAM['L2Banks'])
gpu_info['numSQC'] = str(soc_module.SOC_PARAM['numSQC'])
gpu_info['LDSBanks'] = str(soc_module.SOC_PARAM['LDSBanks'])
gpu_info['numPipes'] = str(soc_module.SOC_PARAM['numPipes'])
except KeyError as e:
error("Incomplete class definition for %s. Expected a field for %s in SOC_PARAM." % (gpu_arch, e))\
@@ -358,6 +362,7 @@ def get_machine_specs(devicenum):
gpu_info['L2Banks'],
gpu_info['LDSBanks'],
gpu_info['numSQC'],
gpu_info['numPipes'],
hbmBW,
compute_partition,
memory_partition,
projects/rocprofiler-compute/src/utils/utils.py
+2 -1
Vedi File
@@ -299,7 +299,7 @@ def gen_sysinfo(workload_name, workload_dir, ip_blocks, app_cmd, skip_roof, roof
header += "command,"
header += "host_name,host_cpu,sbios,host_distro,host_kernel,host_rocmver,date,"
header += "gpu_soc,vbios,numSE,numCU,numSIMD,waveSize,maxWavesPerCU,maxWorkgroupSize,"
header += "L1,L2,sclk,mclk,cur_sclk,cur_mclk,L2Banks,LDSBanks,name,numSQC,hbmBW,compute_partition,memory_partition,"
header += "L1,L2,sclk,mclk,cur_sclk,cur_mclk,L2Banks,LDSBanks,name,numSQC,numPipes,hbmBW,compute_partition,memory_partition,"
header += "ip_blocks\n"
sysinfo.write(header)
@@ -344,6 +344,7 @@ def gen_sysinfo(workload_name, workload_dir, ip_blocks, app_cmd, skip_roof, roof
mspec.LDSBanks,
mspec.GPU,
mspec.numSQC,
mspec.numPipes,
mspec.hbmBW,
mspec.compute_partition,
mspec.memory_partition,